Data Scientist Nanodegree

Convolutional Neural Networks

Project: Write an Algorithm for a Dog Identification App

This notebook walks you through one of the most popular Udacity projects across machine learning and artificial intellegence nanodegree programs. The goal is to classify images of dogs according to their breed.

If you are looking for a more guided capstone project related to deep learning and convolutional neural networks, this might be just it. Notice that even if you follow the notebook to creating your classifier, you must still create a blog post or deploy an application to fulfill the requirements of the capstone project.

Also notice, you may be able to use only parts of this notebook (for example certain coding portions or the data) without completing all parts and still meet all requirements of the capstone project.


In this notebook, some template code has already been provided for you, and you will need to implement additional functionality to successfully complete this project. You will not need to modify the included code beyond what is requested. Sections that begin with '(IMPLEMENTATION)' in the header indicate that the following block of code will require additional functionality which you must provide. Instructions will be provided for each section, and the specifics of the implementation are marked in the code block with a 'TODO' statement. Please be sure to read the instructions carefully!

In addition to implementing code, there will be questions that you must answer which relate to the project and your implementation. Each section where you will answer a question is preceded by a 'Question X' header. Carefully read each question and provide thorough answers in the following text boxes that begin with 'Answer:'. Your project submission will be evaluated based on your answers to each of the questions and the implementation you provide.

Note: Code and Markdown cells can be executed using the Shift + Enter keyboard shortcut. Markdown cells can be edited by double-clicking the cell to enter edit mode.

The rubric contains optional "Stand Out Suggestions" for enhancing the project beyond the minimum requirements. If you decide to pursue the "Stand Out Suggestions", you should include the code in this IPython notebook.


Why We're Here

In this notebook, you will make the first steps towards developing an algorithm that could be used as part of a mobile or web app. At the end of this project, your code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling. The image below displays potential sample output of your finished project (... but we expect that each student's algorithm will behave differently!).

Sample Dog Output

In this real-world setting, you will need to piece together a series of models to perform different tasks; for instance, the algorithm that detects humans in an image will be different from the CNN that infers dog breed. There are many points of possible failure, and no perfect algorithm exists. Your imperfect solution will nonetheless create a fun user experience!

The Road Ahead

We break the notebook into separate steps. Feel free to use the links below to navigate the notebook.

  • Step 0: Import Datasets
  • Step 1: Detect Humans
  • Step 2: Detect Dogs
  • Step 3: Create a CNN to Classify Dog Breeds (from Scratch)
  • Step 4: Use a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)
  • Step 6: Write your Algorithm
  • Step 7: Test Your Algorithm

Step 0: Import Datasets

Import Dog Dataset

In the code cell below, we import a dataset of dog images. We populate a few variables through the use of the load_files function from the scikit-learn library:

  • train_files, valid_files, test_files - numpy arrays containing file paths to images
  • train_targets, valid_targets, test_targets - numpy arrays containing onehot-encoded classification labels
  • dog_names - list of string-valued dog breed names for translating labels
In [1]:
from sklearn.datasets import load_files       
from keras.utils import np_utils
import numpy as np
from glob import glob

# define function to load train, test, and validation datasets
def load_dataset(path):
    data = load_files(path)
    dog_files = np.array(data['filenames'])
    dog_targets = np_utils.to_categorical(np.array(data['target']), 133)
    return dog_files, dog_targets

# load train, test, and validation datasets
train_files, train_targets = load_dataset('../../../data/dog_images/train')
valid_files, valid_targets = load_dataset('../../../data/dog_images/valid')
test_files, test_targets = load_dataset('../../../data/dog_images/test')

# load list of dog names
dog_names = [item[20:-1] for item in sorted(glob("../../../data/dog_images/train/*/"))]

# print statistics about the dataset
print('There are %d total dog categories.' % len(dog_names))
print('There are %s total dog images.\n' % len(np.hstack([train_files, valid_files, test_files])))
print('There are %d training dog images.' % len(train_files))
print('There are %d validation dog images.' % len(valid_files))
print('There are %d test dog images.'% len(test_files))
Using TensorFlow backend.
There are 133 total dog categories.
There are 8351 total dog images.

There are 6680 training dog images.
There are 835 validation dog images.
There are 836 test dog images.

Import Human Dataset

In the code cell below, we import a dataset of human images, where the file paths are stored in the numpy array human_files.

In [2]:
import random
random.seed(8675309)

# load filenames in shuffled human dataset
human_files = np.array(glob("../../../data/lfw/*/*"))
random.shuffle(human_files)

# print statistics about the dataset
print('There are %d total human images.' % len(human_files))
There are 13233 total human images.

Step 1: Detect Humans

We use OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images. OpenCV provides many pre-trained face detectors, stored as XML files on github. We have downloaded one of these detectors and stored it in the haarcascades directory.

In the next code cell, we demonstrate how to use this detector to find human faces in a sample image.

In [3]:
import cv2                
import matplotlib.pyplot as plt                        
%matplotlib inline                               

# extract pre-trained face detector
face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_default.xml')

# load color (BGR) image
img = cv2.imread(human_files[288])
# convert BGR image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# find faces in image
faces = face_cascade.detectMultiScale(gray)

# print number of faces detected in the image
print('Number of faces detected:', len(faces))

# get bounding box for each detected face
for (x,y,w,h) in faces:
    # add bounding box to color image
    cv2.rectangle(img,(x,y),(x+w,y+h),(255,0,0),2)
    
# convert BGR image to RGB for plotting
cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# display the image, along with bounding box
plt.imshow(cv_rgb)
plt.show()
Number of faces detected: 4

Before using any of the face detectors, it is standard procedure to convert the images to grayscale. The detectMultiScale function executes the classifier stored in face_cascade and takes the grayscale image as a parameter.

In the above code, faces is a numpy array of detected faces, where each row corresponds to a detected face. Each detected face is a 1D array with four entries that specifies the bounding box of the detected face. The first two entries in the array (extracted in the above code as x and y) specify the horizontal and vertical positions of the top left corner of the bounding box. The last two entries in the array (extracted here as w and h) specify the width and height of the box.

Write a Human Face Detector

We can use this procedure to write a function that returns True if a human face is detected in an image and False otherwise. This function, aptly named face_detector, takes a string-valued file path to an image as input and appears in the code block below.

In [4]:
# returns "True" if face is detected in image stored at img_path
def face_detector(img_path):
    
    """
        Description:
                Detect whether the human face is detected in an image
        Input :
                img_path : path to image        
        Output:
                boolean whether the human face is detected in an image or not
    """
    
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    
    return len(faces) > 0

(IMPLEMENTATION) Assess the Human Face Detector

Question 1: Use the code cell below to test the performance of the face_detector function.

  • What percentage of the first 100 images in human_files have a detected human face?
  • What percentage of the first 100 images in dog_files have a detected human face?

Ideally, we would like 100% of human images with a detected face and 0% of dog images with a detected face. You will see that our algorithm falls short of this goal, but still gives acceptable performance. We extract the file paths for the first 100 images from each of the datasets and store them in the numpy arrays human_files_short and dog_files_short.

Answer:

  • Look at the printed-result below
In [5]:
human_files_short = human_files[:100]
dog_files_short = train_files[:100]
# Do NOT modify the code above this line.

"""=============================== : TODO : ===================================="""

## Test the performance of the face_detector algorithm on the images in human_files_short and dog_files_short.
Nb_human_face_detected = sum([face_detector(path) for path in human_files_short])
Nb_dog_face_detected= sum([face_detector(path) for path in dog_files_short])

# Print out results
print('Percentage of human_faces detected in the first 100 human_files_short: {}%'.
      format(100*int(Nb_human_face_detected / 100)))
print('Percentage of faces detected in dog_files_short: {}%'.format(int(100*Nb_dog_face_detected / 100)))
Percentage of human_faces detected in the first 100 human_files_short: 98%
Percentage of faces detected in dog_files_short: 58%

Question 2: This algorithmic choice necessitates that we communicate to the user that we accept human images only when they provide a clear view of a face (otherwise, we risk having unneccessarily frustrated users!).

  • In your opinion, is this a reasonable expectation to pose on the user?
  • If not, can you think of a way to detect humans in images that does not necessitate an image with a clearly presented face?

Answer:

We suggest the face detector from OpenCV as a potential way to detect human images in your algorithm, but you are free to explore other approaches, especially approaches that make use of deep learning :). Please use the code cell below to design and test your own face detection algorithm. If you decide to pursue this optional task, report performance on each of the datasets.

In [6]:
import os
print(os.listdir('haarcascades'))
['haarcascade_frontalface_alt2.xml', 'haarcascade_frontalface_alt.xml', '.ipynb_checkpoints', 'haarcascade_frontalface_default.xml', 'haarcascade_frontalface_alt_tree.xml']
In [7]:
img_folder = 'haarcascades' + '/haarcascade_frontalface_default.xml'
cv2.CascadeClassifier(img_folder)
Out[7]:
<CascadeClassifier 0x7f2520365f90>
In [8]:
## (Optional) TODO: Report the performance of another face detection algorithm on the LFW dataset
### Feel free to use as many code cells as needed.

def face_detector2(img_path, xml_file = 'haarcascade_frontalface_default.xml'):
    '''
    Using a CascadeClassifier from cv2 to check whether or not a face of a person is present in an image.
    This Classifier works with the haarcascade-file 'haarcascade_frontalface_alt2.xml'
    Args:
        img_path: path to an image
    Returns:
        Return true, if a face of a person is present on the image. Else return false
    '''
    
    # Here I try another haarcascades classifier from opencv
    # In this case -> haarcascade_frontalface_alt_tree.xml
    
    face_cascade = cv2.CascadeClassifier('haarcascades/haarcascade_frontalface_alt2.xml') 

    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray)
    return len(faces) > 0

Nb_human_face_detected = sum([face_detector2(path) for path in human_files_short])
Nb_dog_face_detected= sum([face_detector2(path) for path in dog_files_short])
print('Percentage of human_faces detected in the first 100 human_files_short: {}%'.
      format(100*int(Nb_human_face_detected / 100)))
print('Percentage of faces detected in dog_files_short: {}%'.format(int(100*Nb_dog_face_detected / 100)))
Percentage of human_faces detected in the first 100 human_files_short: 100%
Percentage of faces detected in dog_files_short: 20%

Step 2: Detect Dogs

In this section, we use a pre-trained ResNet-50 model to detect dogs in images.

  • Our first line of code downloads the ResNet-50 model, along with weights that have been trained on ImageNet, a very large, very popular dataset used for image classification and other vision tasks. ImageNet contains over 10 million URLs, each linking to an image containing an object from one of 1000 categories.
  • Given an image, this pre-trained ResNet-50 model returns a prediction (derived from the available categories in ImageNet) for the object that is contained in the image.
In [9]:
from keras.applications.resnet50 import ResNet50

# define ResNet50 model
ResNet50_model = ResNet50(weights = 'imagenet')
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels.h5
102858752/102853048 [==============================] - 1s 0us/step

Pre-process the Data

When using TensorFlow as backend, Keras CNNs require a 4D array (which we'll also refer to as a 4D tensor) as input, with shape

$$ (\text{nb_samples}, \text{rows}, \text{columns}, \text{channels}), $$

where nb_samples corresponds to the total number of images (or samples), and rows, columns, and channels correspond to the number of rows, columns, and channels for each image, respectively.

The path_to_tensor function below takes a string-valued file path to a color image as input and returns a 4D tensor suitable for supplying to a Keras CNN. The function first loads the image and resizes it to a square image that is $224 \times 224$ pixels. Next, the image is converted to an array, which is then resized to a 4D tensor. In this case, since we are working with color images, each image has three channels. Likewise, since we are processing a single image (or sample), the returned tensor will always have shape

$$ (1, 224, 224, 3). $$

The paths_to_tensor function takes a numpy array of string-valued image paths as input and returns a 4D tensor with shape

$$ (\text{nb_samples}, 224, 224, 3). $$

Here, nb_samples is the number of samples, or number of images, in the supplied array of image paths. It is best to think of nb_samples as the number of 3D tensors (where each 3D tensor corresponds to a different image) in your dataset!

In [10]:
from keras.preprocessing import image                  
from tqdm import tqdm

def path_to_tensor(img_path):
    """
        Description: convert the path_of_image to a tensor 4D
        Input :
            img_path : directory / path to image
        Output:
            a 4D tensor
    """
    # loads RGB image as PIL.Image.Image type
    img = image.load_img(img_path, target_size = (224, 224))
    
    # convert PIL.Image.Image type to 3D tensor with shape (224, 224, 3)
    x = image.img_to_array(img)
    
    # convert 3D tensor to 4D tensor with shape (1, 224, 224, 3) and return 4D tensor
    return np.expand_dims(x, axis=0)

def paths_to_tensor(img_paths):
    """
        Description: load all your image to a tensors as a list
        Input:
            img_path: path to your uploading image
        Output:
            An array that stacked the list of the 4D tensors
    """
    
    list_of_tensors = [path_to_tensor(img_path) for img_path in tqdm(img_paths)]
    
    return np.vstack(list_of_tensors)

Making Predictions with ResNet-50

Getting the 4D tensor ready for ResNet-50, and for any other pre-trained model in Keras, requires some additional processing. First, the RGB image is converted to BGR by reordering the channels. All pre-trained models have the additional normalization step that the mean pixel (expressed in RGB as $$[103.939, 116.779, 123.68]$$ and calculated from all pixels in all images in ImageNet) must be subtracted from every pixel in each image. This is implemented in the imported function preprocess_input. If you're curious, you can check the code for preprocess_input here.

Now that we have a way to format our image for supplying to ResNet-50, we are now ready to use the model to extract the predictions. This is accomplished with the predict method, which returns an array whose $i$-th entry is the model's predicted probability that the image belongs to the $i$-th ImageNet category. This is implemented in the ResNet50_predict_labels function below.

By taking the argmax of the predicted probability vector, we obtain an integer corresponding to the model's predicted object class, which we can identify with an object category through the use of this dictionary.

In [11]:
from keras.applications.resnet50 import preprocess_input, decode_predictions

def ResNet50_predict_labels(img_path):
    """
        Description
            This function print out the indices of the maximum values along an axis of the input array
        Input
            img_path : path to image
        Output
            the indices of the maximum values along an axis of the array of image.
    """
    # returns prediction vector for image located at img_path
    img = preprocess_input(path_to_tensor(img_path))
    
    return np.argmax(ResNet50_model.predict(img))

Write a Dog Detector

  • While looking at the dictionary, you will notice that the categories corresponding to dogs appear in an uninterrupted sequence and correspond to dictionary keys 151-268, inclusive, to include all categories from 'Chihuahua' to 'Mexican hairless'.

  • Thus, in order to check to see if an image is predicted to contain a dog by the pre-trained ResNet-50 model, we need only check if the ResNet50_predict_labels function above returns a value between 151 and 268 (inclusive).

We use these ideas to complete the dog_detector function below, which returns True if a dog is detected in an image (and False if not).

In [12]:
### returns "True" if a dog is detected in the image stored at img_path
def dog_detector(img_path):
    """
        Description: 
            returns "True" if a dog is detected in the image stored at img_path
        Input  
            img_path : path to image
        Output 
            boolean value of detection result 
    """
    prediction = ResNet50_predict_labels(img_path)
    
    return ((prediction <= 268) & (prediction >= 151)) 

Display & verify

In [13]:
files = [human_files_short[0], dog_files_short[0]]
fig, ax = plt.subplots(nrows = 1, ncols = 2, figsize = (20, 5))
for idx, file in enumerate(files):
    load_img = cv2.imread(file)
    rgb_img = cv2.cvtColor(load_img, cv2.COLOR_BGR2RGB)
    ax[idx].imshow(rgb_img)
    ax[idx].set_title('Dog detector result: ' + str(dog_detector(file)))

(IMPLEMENTATION) Assess the Dog Detector

Question 3: Use the code cell below to test the performance of your dog_detector function.

  • What percentage of the images in human_files_short have a detected dog?
  • What percentage of the images in dog_files_short have a detected dog?

Answer:

  • Look at the result bellow
In [14]:
### TODO: Test the performance of the dog_detector function on the images in human_files_short and dog_files_short.

Nb_human_face_detected = sum([dog_detector(path) for path in human_files_short])
Nb_dog_face_detected= sum([dog_detector(path) for path in dog_files_short])

print('Percentage of dogs detected in the first 100 human_files_short: {}%'.
      format(100*int(Nb_human_face_detected / 100)))
print('Percentage of dogs detected in dog_files_short: {}%'.
      format(int(100*Nb_dog_face_detected / 100)))
Percentage of dogs detected in the first 100 human_files_short: 0%
Percentage of dogs detected in dog_files_short: 100%

Step 3: Create a CNN to Classify Dog Breeds (from Scratch)

Now that we have functions for detecting humans and dogs in images, we need a way to predict breed from images. In this step, you will create a CNN that classifies dog breeds. You must create your CNN from scratch (so, you can't use transfer learning yet!), and you must attain a test accuracy of at least 1%. In Step 5 of this notebook, you will have the opportunity to use transfer learning to create a CNN that attains greatly improved accuracy.

In [15]:
from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
from keras.layers import Dropout, Flatten, Dense
from keras.models import Sequential

model = Sequential()

Be careful with adding too many trainable layers! More parameters means longer training, which means you are more likely to need a GPU to accelerate the training process. Thankfully, Keras provides a handy estimate of the time that each epoch is likely to take; you can extrapolate this estimate to figure out how long it will take for your algorithm to train.

We mention that the task of assigning breed to dogs from images is considered exceptionally challenging. To see why, consider that even a human would have great difficulty in distinguishing between a Brittany and a Welsh Springer Spaniel.

Brittany Welsh Springer Spaniel

It is not difficult to find other dog breed pairs with minimal inter-class variation (for instance, Curly-Coated Retrievers and American Water Spaniels).

Curly-Coated Retriever American Water Spaniel

Likewise, recall that labradors come in yellow, chocolate, and black. Your vision-based algorithm will have to conquer this high intra-class variation to determine how to classify all of these different shades as the same breed.

Yellow Labrador Chocolate Labrador Black Labrador

We also mention that random chance presents an exceptionally low bar: setting aside the fact that the classes are slightly imabalanced, a random guess will provide a correct answer roughly 1 in 133 times, which corresponds to an accuracy of less than 1%.

Remember that the practice is far ahead of the theory in deep learning. Experiment with many different architectures, and trust your intuition. And, of course, have fun!

Pre-process the Data

We rescale the images by dividing every pixel in every image by 255.

In [16]:
from PIL import ImageFile                            
ImageFile.LOAD_TRUNCATED_IMAGES = True                 

# pre-process the data for Keras
train_tensors = paths_to_tensor(train_files).astype('float32')/255
valid_tensors = paths_to_tensor(valid_files).astype('float32')/255
test_tensors = paths_to_tensor(test_files).astype('float32')/255
100%|██████████| 6680/6680 [01:34<00:00, 70.61it/s] 
100%|██████████| 835/835 [00:33<00:00, 24.76it/s]
100%|██████████| 836/836 [00:10<00:00, 79.03it/s] 

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    model.summary()

We have imported some Python modules to get you started, but feel free to import as many modules as you need. If you end up getting stuck, here's a hint that specifies a model that trains relatively fast on CPU and attains >1% test accuracy in 5 epochs:

Sample CNN

Question 4: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. If you chose to use the hinted architecture above, describe why you think that CNN architecture should work well for the image classification task.

Answer:

  • Firstly, I keep the idea of the instruction in this section,
  • I used 3 Convolutional Layers with 16, 32, 64 filters respectively in order to discover features from the images.
  • Beside that, between these convolutional layers; added a Max Pooling Layer.
  • I used a Dense Layer with a ReLU (the 1st and 3rd) and sigmoid activation functions at the convolutional layers
  • For the output layer, I used a softmax activation function to get results in percentage.
  • Moreover, I had added the Dropout to prevent overfiting :)
In [17]:
### TODO: Define your architecture.
model.add(Conv2D(16, kernel_size = 2, strides = (1, 1),
                 input_shape = (224, 224, 3),
                 activation = 'relu'))
model.add(MaxPooling2D(pool_size = 2, strides = (2, 2), padding = 'valid'))
model.add(Conv2D(32, kernel_size = 2, 
                 strides = (1, 1), activation = 'sigmoid'))
model.add(MaxPooling2D(pool_size = 2, strides = (2, 2), padding = 'same'))
model.add(Conv2D(64, kernel_size = 2, 
                 strides = (1, 1), activation = 'relu'))
model.add(MaxPooling2D(pool_size = 2, strides = (2, 2), padding = 'same'))
model.add(GlobalAveragePooling2D())
model.add(Dropout(0.25))
model.add(Dense(133, activation = 'softmax'))
model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 223, 223, 16)      208       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 111, 111, 16)      0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 110, 110, 32)      2080      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 55, 55, 32)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 54, 54, 64)        8256      
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 27, 27, 64)        0         
_________________________________________________________________
global_average_pooling2d_1 ( (None, 64)                0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 64)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 133)               8645      
=================================================================
Total params: 19,189
Trainable params: 19,189
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [18]:
model.compile(optimizer = 'rmsprop', 
              loss = 'categorical_crossentropy', 
              metrics=['accuracy'])

Before train the model, I will generate batches of tensor image data with real-time data augmentation by using the preprocessing.image on keras

Reference

Nhớ đổi tên mấy thằng dưới đây

Create and configure augmented image generator

In [19]:
from keras.preprocessing.image import ImageDataGenerator

img_gen_train = ImageDataGenerator(
                        fill_mode = 'constant',  # Points outside the boundaries of the input are filled according to the
                                                 # given mode: kkkkkkkk|abcd|kkkkkkkk (cval=k)
                        width_shift_range = 0.15,  # randomly shift images horizontally (15% of total width)
                        height_shift_range = 0.15,  # randomly shift images vertically (15% of total height)
                        horizontal_flip = True) # randomly flip images horizontally

img_gen_valid = ImageDataGenerator(
                        fill_mode = 'constant', # Points outside the boundaries of the input are filled according to the 
                                                # given mode: kkkkkkkk|abcd|kkkkkkkk (cval=k)
                        width_shift_range = 0.15,  # randomly shift images horizontally (15% of total width)
                        height_shift_range = 0.15,  # randomly shift images vertically (15% of total height)
                        horizontal_flip = True) # randomly flip images horizontally

img_gen_train
Out[19]:
<keras.preprocessing.image.ImageDataGenerator at 0x7f24cd74d710>

Fit augmented image generator on data

In [20]:
img_gen_train.fit(train_tensors)
img_gen_valid.fit(valid_tensors)

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [21]:
from keras.callbacks import ModelCheckpoint  

### TODO: specify the number of epochs that you would like to use to train the model.
epochs = 3
batch_size = 64 # I have tried 16, 32 before but not good

### Do NOT modify the code below this line.
checkpointer = ModelCheckpoint(filepath = 'saved_models/weights.best.from_scratch.hdf5', 
                               verbose = 2, 
                               save_best_only = True)

history = model.fit_generator(
                    img_gen_train.flow(train_tensors, train_targets,        # training data (X, y, batchsize)
                                       batch_size = batch_size),
                    steps_per_epoch = train_tensors.shape[0] // batch_size, # Total number of steps (batches of samples) before declaring 
                                                                            # one epoch finished and starting the next epoch. 
                    epochs = epochs, # Number of samples per gradient update.
                    verbose = 1,     # Verbosity mode
                    callbacks = [checkpointer],
                    validation_data = img_gen_valid.flow(valid_tensors, valid_targets,# Data on which to evaluate the loss and any model metrics 
                                                                                    # at the end of each epoch. 
                                                         batch_size = batch_size), # The model will not be trained on this data.
                    validation_steps = valid_tensors.shape[0] // batch_size  #  Total number of steps (batches of samples) to draw before 
                                                                             # stopping when performing validation at the end of every epoch.
                   )
Epoch 1/3
103/104 [============================>.] - ETA: 5s - loss: 4.8864 - acc: 0.0083 Epoch 00001: val_loss improved from inf to 4.87419, saving model to saved_models/weights.best.from_scratch.hdf5
104/104 [==============================] - 588s 6s/step - loss: 4.8869 - acc: 0.0083 - val_loss: 4.8742 - val_acc: 0.0132
Epoch 2/3
103/104 [============================>.] - ETA: 5s - loss: 4.8730 - acc: 0.0130 Epoch 00002: val_loss improved from 4.87419 to 4.85726, saving model to saved_models/weights.best.from_scratch.hdf5
104/104 [==============================] - 557s 5s/step - loss: 4.8731 - acc: 0.0131 - val_loss: 4.8573 - val_acc: 0.0096
Epoch 3/3
103/104 [============================>.] - ETA: 5s - loss: 4.8589 - acc: 0.0132 Epoch 00003: val_loss improved from 4.85726 to 4.84222, saving model to saved_models/weights.best.from_scratch.hdf5
104/104 [==============================] - 558s 5s/step - loss: 4.8587 - acc: 0.0131 - val_loss: 4.8422 - val_acc: 0.0120

Load the Model with the Best Validation Loss

In [22]:
model.load_weights('saved_models/weights.best.from_scratch.hdf5')

Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 1%.

In [24]:
# get index of predicted dog breed for each image in test set
dog_breed_predictions = [np.argmax(model.predict(np.expand_dims(tensor, axis=0))) for tensor in test_tensors]

# report test accuracy
test_accuracy = 100*np.sum(np.array(dog_breed_predictions)==np.argmax(test_targets, axis=1))/len(dog_breed_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 1.5703%

Summarize history for loss and accuracy on both training and validattion sets

In [41]:
def display_model_acc_loss(model_history = history, loc = 'best', figsize = (20, 8)):
    fig, ax = plt.subplots(nrows = 1, ncols = 2, figsize = figsize)
    plot_types = ['loss', 'acc']
    titles = ['loss', 'accuracy']
    for k in range(2):
        ax[k].plot(model_history.history[plot_types[k]])
        ax[k].plot(model_history.history['val_' + plot_types[k]])
        ax[k].set_ylabel(titles[k])
        ax[k].set_xlabel('epoch')
        ax[k].legend(['train', 'validation'], loc = loc)
        ax[k].set_title(titles[k])
    plt.show()
    
display_model_acc_loss(history)

Step 4: Use a CNN to Classify Dog Breeds

To reduce training time without sacrificing accuracy, we show you how to train a CNN using transfer learning. In the following step, you will get a chance to use transfer learning to train your own CNN.

Obtain Bottleneck Features

In [34]:
bottleneck_features = np.load('bottleneck_features/DogVGG16Data.npz')
train_VGG16 = bottleneck_features['train']
valid_VGG16 = bottleneck_features['valid']
test_VGG16 = bottleneck_features['test']

Model Architecture

The model uses the the pre-trained VGG-16 model as a fixed feature extractor, where the last convolutional output of VGG-16 is fed as input to our model. We only add a global average pooling layer and a fully connected layer, where the latter contains one node for each dog category and is equipped with a softmax.

In [35]:
VGG16_model = Sequential()
VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
# But here, i want to check what will happen if adding the drop-out
VGG16_model.add(Dropout(0.1))
VGG16_model.add(Dense(133, activation='softmax'))

VGG16_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_2 ( (None, 512)               0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 133)               68229     
=================================================================
Total params: 68,229
Trainable params: 68,229
Non-trainable params: 0
_________________________________________________________________

Compile the Model

In [36]:
VGG16_model.compile(loss = 'categorical_crossentropy', 
                    optimizer = 'rmsprop',
                    metrics = ['accuracy'])

Train the Model

In [37]:
checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.VGG16.hdf5', 
                               verbose = 1, save_best_only = True)

VGG16_history = VGG16_model.fit(train_VGG16, train_targets, 
                                validation_data=(valid_VGG16, valid_targets),
                                epochs = 20, batch_size = 32, 
                                callbacks=[checkpointer], verbose = 1)
Train on 6680 samples, validate on 835 samples
Epoch 1/20
6528/6680 [============================>.] - ETA: 0s - loss: 12.9923 - acc: 0.0991Epoch 00001: val_loss improved from inf to 11.32868, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 4s 574us/step - loss: 12.9661 - acc: 0.1001 - val_loss: 11.3287 - val_acc: 0.1832
Epoch 2/20
6400/6680 [===========================>..] - ETA: 0s - loss: 10.7285 - acc: 0.2234Epoch 00002: val_loss improved from 11.32868 to 10.07166, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 192us/step - loss: 10.7249 - acc: 0.2243 - val_loss: 10.0717 - val_acc: 0.2766
Epoch 3/20
6624/6680 [============================>.] - ETA: 0s - loss: 9.9020 - acc: 0.2951Epoch 00003: val_loss improved from 10.07166 to 9.57544, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 196us/step - loss: 9.8973 - acc: 0.2954 - val_loss: 9.5754 - val_acc: 0.3246
Epoch 4/20
6656/6680 [============================>.] - ETA: 0s - loss: 9.3147 - acc: 0.3478Epoch 00004: val_loss improved from 9.57544 to 9.23721, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 195us/step - loss: 9.3119 - acc: 0.3478 - val_loss: 9.2372 - val_acc: 0.3461
Epoch 5/20
6624/6680 [============================>.] - ETA: 0s - loss: 8.9832 - acc: 0.3777Epoch 00005: val_loss improved from 9.23721 to 9.01643, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 179us/step - loss: 8.9756 - acc: 0.3780 - val_loss: 9.0164 - val_acc: 0.3689
Epoch 6/20
6528/6680 [============================>.] - ETA: 0s - loss: 8.7084 - acc: 0.4001Epoch 00006: val_loss improved from 9.01643 to 8.83224, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 191us/step - loss: 8.7161 - acc: 0.3999 - val_loss: 8.8322 - val_acc: 0.3737
Epoch 7/20
6592/6680 [============================>.] - ETA: 0s - loss: 8.5449 - acc: 0.4191Epoch 00007: val_loss improved from 8.83224 to 8.75301, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 201us/step - loss: 8.5332 - acc: 0.4196 - val_loss: 8.7530 - val_acc: 0.3928
Epoch 8/20
6496/6680 [============================>.] - ETA: 0s - loss: 8.4066 - acc: 0.4373Epoch 00008: val_loss improved from 8.75301 to 8.72407, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 203us/step - loss: 8.4266 - acc: 0.4365 - val_loss: 8.7241 - val_acc: 0.3880
Epoch 9/20
6464/6680 [============================>.] - ETA: 0s - loss: 8.3351 - acc: 0.4437Epoch 00009: val_loss improved from 8.72407 to 8.67612, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 205us/step - loss: 8.3402 - acc: 0.4430 - val_loss: 8.6761 - val_acc: 0.3952
Epoch 10/20
6592/6680 [============================>.] - ETA: 0s - loss: 8.2249 - acc: 0.4481Epoch 00010: val_loss improved from 8.67612 to 8.50525, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 206us/step - loss: 8.2305 - acc: 0.4476 - val_loss: 8.5052 - val_acc: 0.4036
Epoch 11/20
6592/6680 [============================>.] - ETA: 0s - loss: 8.0978 - acc: 0.4660Epoch 00011: val_loss improved from 8.50525 to 8.44757, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 201us/step - loss: 8.0981 - acc: 0.4659 - val_loss: 8.4476 - val_acc: 0.4024
Epoch 12/20
6400/6680 [===========================>..] - ETA: 0s - loss: 8.0408 - acc: 0.4689Epoch 00012: val_loss improved from 8.44757 to 8.35762, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 195us/step - loss: 8.0409 - acc: 0.4696 - val_loss: 8.3576 - val_acc: 0.4144
Epoch 13/20
6656/6680 [============================>.] - ETA: 0s - loss: 7.9787 - acc: 0.4703Epoch 00013: val_loss improved from 8.35762 to 8.30304, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 187us/step - loss: 7.9743 - acc: 0.4707 - val_loss: 8.3030 - val_acc: 0.4132
Epoch 14/20
6496/6680 [============================>.] - ETA: 0s - loss: 7.8369 - acc: 0.4804Epoch 00014: val_loss improved from 8.30304 to 8.21617, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 206us/step - loss: 7.8475 - acc: 0.4801 - val_loss: 8.2162 - val_acc: 0.4216
Epoch 15/20
6528/6680 [============================>.] - ETA: 0s - loss: 7.8035 - acc: 0.4867Epoch 00015: val_loss improved from 8.21617 to 8.20578, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 189us/step - loss: 7.8082 - acc: 0.4867 - val_loss: 8.2058 - val_acc: 0.4251
Epoch 16/20
6368/6680 [===========================>..] - ETA: 0s - loss: 7.7333 - acc: 0.4943Epoch 00016: val_loss improved from 8.20578 to 8.17565, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 195us/step - loss: 7.7524 - acc: 0.4927 - val_loss: 8.1756 - val_acc: 0.4240
Epoch 17/20
6496/6680 [============================>.] - ETA: 0s - loss: 7.7127 - acc: 0.4952Epoch 00017: val_loss did not improve
6680/6680 [==============================] - 1s 181us/step - loss: 7.7115 - acc: 0.4955 - val_loss: 8.2011 - val_acc: 0.4216
Epoch 18/20
6400/6680 [===========================>..] - ETA: 0s - loss: 7.7114 - acc: 0.4998Epoch 00018: val_loss improved from 8.17565 to 8.14209, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 184us/step - loss: 7.6873 - acc: 0.5015 - val_loss: 8.1421 - val_acc: 0.4251
Epoch 19/20
6400/6680 [===========================>..] - ETA: 0s - loss: 7.5980 - acc: 0.5053Epoch 00019: val_loss improved from 8.14209 to 8.02684, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 192us/step - loss: 7.5920 - acc: 0.5057 - val_loss: 8.0268 - val_acc: 0.4275
Epoch 20/20
6560/6680 [============================>.] - ETA: 0s - loss: 7.4838 - acc: 0.5110Epoch 00020: val_loss improved from 8.02684 to 7.92551, saving model to saved_models/weights.best.VGG16.hdf5
6680/6680 [==============================] - 1s 206us/step - loss: 7.4871 - acc: 0.5109 - val_loss: 7.9255 - val_acc: 0.4251

Load the Model with the Best Validation Loss

In [38]:
VGG16_model.load_weights('saved_models/weights.best.VGG16.hdf5')

Summarize history for loss and accuracy

In [43]:
%time display_model_acc_loss(VGG16_history)
CPU times: user 575 ms, sys: 0 ns, total: 575 ms
Wall time: 599 ms

Test the Model

Now, we can use the CNN to test how well it identifies breed within our test dataset of dog images. We print the test accuracy below.

In [44]:
# get index of predicted dog breed for each image in test set
VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis = 0))) for feature in test_VGG16]

# report test accuracy
test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis = 1))/len(VGG16_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 42.4641%

Predict Dog Breed with the Model

In [45]:
from extract_bottleneck_features import *

def VGG16_predict_breed(img_path):
    
    # extract bottleneck features
    bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
    
    # obtain predicted vector
    predicted_vector = VGG16_model.predict(bottleneck_feature)
    
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Verify

In [47]:
image_paths = dog_files_short[:9]

fig, ax = plt.subplots(nrows = 3, ncols = 3, figsize = (20, 20))
ax = ax.ravel()

for k in range(9):
    images = cv2.imread(image_paths[k])
    img_rgb = cv2.cvtColor(images, cv2.COLOR_BGR2RGB)
    ax[k].imshow(img_rgb)
    dog_breed = VGG16_predict_breed(image_paths[k]).replace('ages/train/', '')[4:]
    ax[k].set_title('predict result:' + dog_breed)

Step 5: Create a CNN to Classify Dog Breeds (using Transfer Learning)

You will now use transfer learning to create a CNN that can identify dog breed from images. Your CNN must attain at least 60% accuracy on the test set.

In Step 4, we used transfer learning to create a CNN using VGG-16 bottleneck features. In this section, you must use the bottleneck features from a different pre-trained model. To make things easier for you, we have pre-computed the features for all of the networks that are currently available in Keras:

In [48]:
os.listdir('bottleneck_features')
Out[48]:
['DogVGG16Data.npz', '.gitignore', 'DogResnet50Data.npz']

The files are encoded as such:

Dog{network}Data.npz

where {network}, in the above filename, can be one of VGG19, Resnet50, InceptionV3, or Xception. Pick one of the above architectures, download the corresponding bottleneck features, and store the downloaded file in the bottleneck_features/ folder in the repository.

(IMPLEMENTATION) Obtain Bottleneck Features

In the code block below, extract the bottleneck features corresponding to the train, test, and validation sets by running the following:

bottleneck_features = np.load('bottleneck_features/Dog{network}Data.npz')
train_{network} = bottleneck_features['train']
valid_{network} = bottleneck_features['valid']
test_{network} = bottleneck_features['test']
In [49]:
### TODO: Obtain bottleneck features from another pre-trained CNN.
bottleneck_features = np.load('bottleneck_features/DogResnet50Data.npz')
train_res50net = bottleneck_features['train']
valid_res50net = bottleneck_features['valid']
test_res50net = bottleneck_features['test']

(IMPLEMENTATION) Model Architecture

Create a CNN to classify dog breed. At the end of your code cell block, summarize the layers of your model by executing the line:

    <your model's name>.summary()

Question 5: Outline the steps you took to get to your final CNN architecture and your reasoning at each step. Describe why you think the architecture is suitable for the current problem.

Answer:

  • In this case, I used the ResNet50 model and I get the accuracy about 84.9%

  • I also added a fully connected layer with 512 nodes and a ReLU activation function to detect more patterns and a Dropout to avoid overfitting.

In [50]:
### TODO: Define your architecture.
res50net_model = Sequential()
res50net_model.add(GlobalAveragePooling2D(input_shape = train_res50net.shape[1:]))
res50net_model.add(Dense(512, activation='relu'))
res50net_model.add(Dropout(0.2))
res50net_model.add(Dense(133, activation='softmax'))

# <your model's name>.summary()
res50net_model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
global_average_pooling2d_3 ( (None, 2048)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 512)               1049088   
_________________________________________________________________
dropout_3 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_4 (Dense)              (None, 133)               68229     
=================================================================
Total params: 1,117,317
Trainable params: 1,117,317
Non-trainable params: 0
_________________________________________________________________

(IMPLEMENTATION) Compile the Model

In [51]:
### TODO: Compile the model.
from keras.optimizers import SGD

res50net_model.compile(loss = 'categorical_crossentropy', 
                       optimizer = SGD(lr = 0.001), metrics = ['accuracy'])

(IMPLEMENTATION) Train the Model

Train your model in the code cell below. Use model checkpointing to save the model that attains the best validation loss.

You are welcome to augment the training data, but this is not a requirement.

In [53]:
### TODO: Train the model.
from keras.preprocessing.image import ImageDataGenerator

checkpointer = ModelCheckpoint(filepath='saved_models/weights.best.Resnet50.hdf5', 
                               verbose = 2, save_best_only = True)

resnet50_history = res50net_model.fit(train_res50net, train_targets, 
                                      validation_data = (valid_res50net, valid_targets),
                                      epochs = 200, batch_size = 64,
                                      callbacks=[checkpointer], verbose = 0)
Epoch 00001: val_loss improved from inf to 0.91699, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00002: val_loss improved from 0.91699 to 0.90053, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00003: val_loss improved from 0.90053 to 0.88738, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00004: val_loss improved from 0.88738 to 0.87427, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00005: val_loss improved from 0.87427 to 0.86093, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00006: val_loss improved from 0.86093 to 0.85213, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00007: val_loss improved from 0.85213 to 0.84043, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00008: val_loss improved from 0.84043 to 0.82965, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00009: val_loss improved from 0.82965 to 0.81858, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00010: val_loss improved from 0.81858 to 0.80944, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00011: val_loss improved from 0.80944 to 0.80003, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00012: val_loss improved from 0.80003 to 0.79163, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00013: val_loss improved from 0.79163 to 0.78281, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00014: val_loss improved from 0.78281 to 0.77664, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00015: val_loss improved from 0.77664 to 0.77028, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00016: val_loss improved from 0.77028 to 0.76295, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00017: val_loss improved from 0.76295 to 0.75598, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00018: val_loss improved from 0.75598 to 0.74789, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00019: val_loss improved from 0.74789 to 0.74222, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00020: val_loss improved from 0.74222 to 0.73761, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00021: val_loss improved from 0.73761 to 0.73254, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00022: val_loss improved from 0.73254 to 0.72409, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00023: val_loss improved from 0.72409 to 0.71943, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00024: val_loss improved from 0.71943 to 0.71520, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00025: val_loss improved from 0.71520 to 0.70995, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00026: val_loss improved from 0.70995 to 0.70644, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00027: val_loss improved from 0.70644 to 0.70239, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00028: val_loss improved from 0.70239 to 0.69698, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00029: val_loss improved from 0.69698 to 0.69378, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00030: val_loss improved from 0.69378 to 0.69077, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00031: val_loss improved from 0.69077 to 0.68825, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00032: val_loss improved from 0.68825 to 0.68515, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00033: val_loss improved from 0.68515 to 0.68052, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00034: val_loss improved from 0.68052 to 0.67590, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00035: val_loss improved from 0.67590 to 0.67142, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00036: val_loss improved from 0.67142 to 0.66894, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00037: val_loss improved from 0.66894 to 0.66383, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00038: val_loss improved from 0.66383 to 0.66162, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00039: val_loss improved from 0.66162 to 0.65862, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00040: val_loss improved from 0.65862 to 0.65493, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00041: val_loss improved from 0.65493 to 0.65241, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00042: val_loss improved from 0.65241 to 0.65062, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00043: val_loss improved from 0.65062 to 0.64781, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00044: val_loss improved from 0.64781 to 0.64585, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00045: val_loss improved from 0.64585 to 0.64341, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00046: val_loss improved from 0.64341 to 0.64022, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00047: val_loss improved from 0.64022 to 0.63957, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00048: val_loss improved from 0.63957 to 0.63682, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00049: val_loss improved from 0.63682 to 0.63448, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00050: val_loss improved from 0.63448 to 0.63191, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00051: val_loss improved from 0.63191 to 0.62830, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00052: val_loss improved from 0.62830 to 0.62774, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00053: val_loss improved from 0.62774 to 0.62562, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00054: val_loss improved from 0.62562 to 0.62366, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00055: val_loss improved from 0.62366 to 0.62223, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00056: val_loss improved from 0.62223 to 0.62170, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00057: val_loss improved from 0.62170 to 0.61821, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00058: val_loss improved from 0.61821 to 0.61659, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00059: val_loss improved from 0.61659 to 0.61419, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00060: val_loss improved from 0.61419 to 0.61266, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00061: val_loss improved from 0.61266 to 0.61198, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00062: val_loss improved from 0.61198 to 0.61006, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00063: val_loss improved from 0.61006 to 0.60732, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00064: val_loss improved from 0.60732 to 0.60639, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00065: val_loss improved from 0.60639 to 0.60352, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00066: val_loss improved from 0.60352 to 0.60206, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00067: val_loss improved from 0.60206 to 0.59946, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00068: val_loss improved from 0.59946 to 0.59782, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00069: val_loss improved from 0.59782 to 0.59720, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00070: val_loss improved from 0.59720 to 0.59714, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00071: val_loss improved from 0.59714 to 0.59575, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00072: val_loss improved from 0.59575 to 0.59475, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00073: val_loss improved from 0.59475 to 0.59328, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00074: val_loss improved from 0.59328 to 0.59311, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00075: val_loss improved from 0.59311 to 0.59257, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00076: val_loss improved from 0.59257 to 0.58934, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00077: val_loss improved from 0.58934 to 0.58800, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00078: val_loss improved from 0.58800 to 0.58594, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00079: val_loss improved from 0.58594 to 0.58580, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00080: val_loss improved from 0.58580 to 0.58452, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00081: val_loss did not improve
Epoch 00082: val_loss improved from 0.58452 to 0.58304, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00083: val_loss improved from 0.58304 to 0.58200, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00084: val_loss improved from 0.58200 to 0.58145, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00085: val_loss improved from 0.58145 to 0.58054, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00086: val_loss improved from 0.58054 to 0.57954, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00087: val_loss improved from 0.57954 to 0.57851, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00088: val_loss improved from 0.57851 to 0.57790, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00089: val_loss improved from 0.57790 to 0.57618, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00090: val_loss improved from 0.57618 to 0.57540, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00091: val_loss improved from 0.57540 to 0.57382, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00092: val_loss did not improve
Epoch 00093: val_loss did not improve
Epoch 00094: val_loss improved from 0.57382 to 0.57245, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00095: val_loss improved from 0.57245 to 0.57005, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00096: val_loss improved from 0.57005 to 0.56963, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00097: val_loss improved from 0.56963 to 0.56932, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00098: val_loss improved from 0.56932 to 0.56835, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00099: val_loss improved from 0.56835 to 0.56753, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00100: val_loss improved from 0.56753 to 0.56678, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00101: val_loss improved from 0.56678 to 0.56519, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00102: val_loss improved from 0.56519 to 0.56477, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00103: val_loss did not improve
Epoch 00104: val_loss improved from 0.56477 to 0.56471, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00105: val_loss improved from 0.56471 to 0.56406, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00106: val_loss improved from 0.56406 to 0.56383, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00107: val_loss improved from 0.56383 to 0.56301, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00108: val_loss improved from 0.56301 to 0.56104, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00109: val_loss improved from 0.56104 to 0.56074, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00110: val_loss improved from 0.56074 to 0.55978, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00111: val_loss improved from 0.55978 to 0.55947, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00112: val_loss improved from 0.55947 to 0.55834, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00113: val_loss improved from 0.55834 to 0.55739, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00114: val_loss improved from 0.55739 to 0.55634, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00115: val_loss did not improve
Epoch 00116: val_loss improved from 0.55634 to 0.55633, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00117: val_loss did not improve
Epoch 00118: val_loss improved from 0.55633 to 0.55587, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00119: val_loss improved from 0.55587 to 0.55503, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00120: val_loss improved from 0.55503 to 0.55426, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00121: val_loss improved from 0.55426 to 0.55353, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00122: val_loss improved from 0.55353 to 0.55157, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00123: val_loss improved from 0.55157 to 0.55041, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00124: val_loss improved from 0.55041 to 0.54959, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00125: val_loss did not improve
Epoch 00126: val_loss improved from 0.54959 to 0.54922, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00127: val_loss did not improve
Epoch 00128: val_loss improved from 0.54922 to 0.54799, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00129: val_loss did not improve
Epoch 00130: val_loss improved from 0.54799 to 0.54779, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00131: val_loss did not improve
Epoch 00132: val_loss improved from 0.54779 to 0.54760, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00133: val_loss improved from 0.54760 to 0.54548, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00134: val_loss improved from 0.54548 to 0.54453, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00135: val_loss did not improve
Epoch 00136: val_loss did not improve
Epoch 00137: val_loss improved from 0.54453 to 0.54416, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00138: val_loss did not improve
Epoch 00139: val_loss did not improve
Epoch 00140: val_loss improved from 0.54416 to 0.54304, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00141: val_loss improved from 0.54304 to 0.54233, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00142: val_loss improved from 0.54233 to 0.54190, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00143: val_loss improved from 0.54190 to 0.54090, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00144: val_loss improved from 0.54090 to 0.53988, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00145: val_loss improved from 0.53988 to 0.53970, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00146: val_loss improved from 0.53970 to 0.53959, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00147: val_loss improved from 0.53959 to 0.53946, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00148: val_loss did not improve
Epoch 00149: val_loss improved from 0.53946 to 0.53866, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00150: val_loss did not improve
Epoch 00151: val_loss did not improve
Epoch 00152: val_loss did not improve
Epoch 00153: val_loss improved from 0.53866 to 0.53709, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00154: val_loss improved from 0.53709 to 0.53595, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00155: val_loss did not improve
Epoch 00156: val_loss did not improve
Epoch 00157: val_loss improved from 0.53595 to 0.53545, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00158: val_loss improved from 0.53545 to 0.53533, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00159: val_loss improved from 0.53533 to 0.53504, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00160: val_loss improved from 0.53504 to 0.53421, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00161: val_loss improved from 0.53421 to 0.53406, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00162: val_loss did not improve
Epoch 00163: val_loss did not improve
Epoch 00164: val_loss did not improve
Epoch 00165: val_loss improved from 0.53406 to 0.53353, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00166: val_loss did not improve
Epoch 00167: val_loss did not improve
Epoch 00168: val_loss improved from 0.53353 to 0.53325, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00169: val_loss improved from 0.53325 to 0.53284, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00170: val_loss improved from 0.53284 to 0.53168, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00171: val_loss improved from 0.53168 to 0.53073, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00172: val_loss did not improve
Epoch 00173: val_loss did not improve
Epoch 00174: val_loss did not improve
Epoch 00175: val_loss improved from 0.53073 to 0.53021, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00176: val_loss did not improve
Epoch 00177: val_loss did not improve
Epoch 00178: val_loss improved from 0.53021 to 0.52948, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00179: val_loss did not improve
Epoch 00180: val_loss did not improve
Epoch 00181: val_loss did not improve
Epoch 00182: val_loss did not improve
Epoch 00183: val_loss improved from 0.52948 to 0.52930, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00184: val_loss improved from 0.52930 to 0.52791, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00185: val_loss improved from 0.52791 to 0.52770, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00186: val_loss did not improve
Epoch 00187: val_loss did not improve
Epoch 00188: val_loss did not improve
Epoch 00189: val_loss improved from 0.52770 to 0.52670, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00190: val_loss did not improve
Epoch 00191: val_loss did not improve
Epoch 00192: val_loss did not improve
Epoch 00193: val_loss improved from 0.52670 to 0.52664, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00194: val_loss improved from 0.52664 to 0.52574, saving model to saved_models/weights.best.Resnet50.hdf5
Epoch 00195: val_loss did not improve
Epoch 00196: val_loss did not improve
Epoch 00197: val_loss did not improve
Epoch 00198: val_loss did not improve
Epoch 00199: val_loss did not improve
Epoch 00200: val_loss did not improve

Summarize history for loss and accuracy

In [54]:
%time display_model_acc_loss(resnet50_history)
CPU times: user 562 ms, sys: 0 ns, total: 562 ms
Wall time: 582 ms

(IMPLEMENTATION) Load the Model with the Best Validation Loss

In [55]:
### TODO: Load the model weights with the best validation loss.
res50net_model.load_weights('saved_models/weights.best.Resnet50.hdf5')

(IMPLEMENTATION) Test the Model

Try out your model on the test dataset of dog images. Ensure that your test accuracy is greater than 60%.

In [57]:
### TODO: Calculate classification accuracy on the test dataset.
# get index of predicted dog breed for each image in test set
resnet50_predictions = [np.argmax(res50net_model.predict(np.expand_dims(feature, axis=0))) for feature in test_res50net]

# report test accuracy
test_accuracy = 100*np.sum(np.array(resnet50_predictions) == np.argmax(test_targets, axis=1))/len(resnet50_predictions)
print('Test accuracy: %.4f%%' % test_accuracy)
Test accuracy: 84.9421%

(IMPLEMENTATION) Predict Dog Breed with the Model

Write a function that takes an image path as input and returns the dog breed (Affenpinscher, Afghan_hound, etc) that is predicted by your model.

Similar to the analogous function in Step 5, your function should have three steps:

  1. Extract the bottleneck features corresponding to the chosen CNN model.
  2. Supply the bottleneck features as input to the model to return the predicted vector. Note that the argmax of this prediction vector gives the index of the predicted dog breed.
  3. Use the dog_names array defined in Step 0 of this notebook to return the corresponding breed.

The functions to extract the bottleneck features can be found in extract_bottleneck_features.py, and they have been imported in an earlier code cell. To obtain the bottleneck features corresponding to your chosen CNN architecture, you need to use the function

extract_{network}

where {network}, in the above filename, should be one of VGG19, Resnet50, InceptionV3, or Xception.

In [58]:
### TODO: Write a function that takes a path to an image as input and returns the dog breed that is predicted by the model.
def ResNet50_predict_breed(img_path):
    
    # extract bottleneck features
    bottleneck_feature = extract_Resnet50(path_to_tensor(img_path))

    # obtain predicted vector
    predicted_vector = res50net_model.predict(bottleneck_feature)
    
    # return dog breed that is predicted by the model
    return dog_names[np.argmax(predicted_vector)]

Display & verify

In [60]:
image_paths = dog_files_short[:9]

fig, ax = plt.subplots(nrows = 3, ncols = 3, figsize = (20, 20))
ax = ax.ravel()

for k in range(9):
    images = cv2.imread(image_paths[k])
    img_rgb = cv2.cvtColor(images, cv2.COLOR_BGR2RGB)
    ax[k].imshow(img_rgb)
    dog_breed = ResNet50_predict_breed(image_paths[k]).replace('ages/train/', '')[4:]
    ax[k].set_title('predicted result: ' + dog_breed)

Step 6: Write your Algorithm

Write an algorithm that accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

  • if a dog is detected in the image, return the predicted breed.
  • if a human is detected in the image, return the resembling dog breed.
  • if neither is detected in the image, provide output that indicates an error.

You are welcome to write your own functions for detecting humans and dogs in images, but feel free to use the face_detector and dog_detector functions developed above. You are required to use your CNN from Step 5 to predict dog breed.

A sample image and output for our algorithm is provided below, but feel free to design your own user experience!

Sample Human Output

This photo looks like an Afghan Hound.

(IMPLEMENTATION) Write your Algorithm

In [71]:
### TODO: Write your algorithm.
### Feel free to use as many code cells as needed.

def predict_breed(img_path, img_width = 10):
    """
        Description: This function is used to determine whether the image contains a human, dog, both or neither
        
        Input
            img_path : path to the image
            img_width (int): width of the displayed image belong, default = 10
            
        Output
            show the image and the comment of detection result
    """
    
    img = cv2.imread(img_path)
    cv_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    ratio = img.shape[0] / img.shape[1]
    fig, ax = plt.subplots(1, 1, figsize = (img_width, int(img_width*ratio)))
    ax.imshow(cv_rgb)
    ax.set_title('image_path is {}'.format(img_path))
    plt.show()
    
    breed = ResNet50_predict_breed(img_path).replace('ages/train/', '')[4:]
    
    if dog_detector(img_path) & (face_detector(img_path) == 0):
        print("Dog detector: True")
        return print("This dog breed can be a {}".format(breed))
        
    elif face_detector(img_path) & (dog_detector(img_path) == 0):
        print("Human detector: True")
        return print("If this one were a dog, he / she would be a ... {}!!".format(breed))
    
    elif face_detector(img_path) & (dog_detector(img_path) == 1):
        print("Dog and human were detected in this image")
        return print("If this one were a dog, he / she would be a ... {}!!".format(breed))
        
    else:
        return print("Oopss... The algoritm must be improved or 'No human or dog has been detected in this image'.")
    
predict_breed('images/sample_human_2.png')
Human detector: True
If this one were a dog, he / she would be a ... Dachshund!!

Step 7: Test Your Algorithm

In this section, you will take your new algorithm for a spin! What kind of dog does the algorithm think that you look like? If you have a dog, does it predict your dog's breed accurately? If you have a cat, does it mistakenly think that your cat is a dog?

(IMPLEMENTATION) Test Your Algorithm on Sample Images!

Test your algorithm at least six images on your computer. Feel free to use any images you like. Use at least two human and two dog images.

Question 6: Is the output better than you expected :) ? Or worse :( ? Provide at least three possible points of improvement for your algorithm.

Answer:

  • Overall, the algorithm is working good as I expected; but the output could be improved better if
  • We used the function face_detector2 to replace face_detector
  • We used Image Augmentation
  • Prevent overfiting by using Dropout better (for example Dropout(0.3) )
  • We have more images for training
  • Increasing the depth of the neural network
  • Firstly, examine what we have in the folder images
In [72]:
## TODO: Execute your algorithm from Step 6 on at least 6 images on your computer.
## Feel free to use as many code cells as needed.
image_list = os.listdir('images')
image_list = [img for img in image_list if img not in ['ipynb_checkpoints', '.ipynb_checkpoints',
                                                       'sample_cnn.png', 'sample_human_2.png']]
image_list = sorted(image_list)
print(image_list)
len(image_list)
['American_water_spaniel_00648.jpg', 'Brittany_02625.jpg', 'Curly-coated_retriever_03896.jpg', 'Labrador_retriever_06449.jpg', 'Labrador_retriever_06455.jpg', 'Labrador_retriever_06457.jpg', 'Welsh_springer_spaniel_08203.jpg', 'dog_1.png', 'human_dog.jpg', 'human_dog_1', 'human_dog_2', 'mess_app_dog.jpg', 'mess_app_dog_2.jpg', 'sample_dog_output.png', 'sample_human_output.png', 'xijingdog.jpg']
Out[72]:
16
  • Pic 1.
In [73]:
image_path = 'images/' + image_list[0]
predict_breed(image_path)
Dog detector: True
This dog breed can be a Boykin_spaniel
  • Pic 2.
In [74]:
image_path = 'images/' + image_list[1]
predict_breed(image_path)
Dog and human were detected in this image
If this one were a dog, he / she would be a ... Brittany!!

=> Cons 1. In this case, the only the dog was in the picture but my algorithm has detected the human and dog! This may be I has use the function face_detector instead of face_detector2-which will better the first one.

  • Pic 3.
In [75]:
image_path = 'images/' + image_list[10]
predict_breed(image_path)
Oopss... The algoritm must be improved or 'No human or dog has been detected in this image'.

=> Cons 2. In this case, the woman and her dog can not be detected. Because the algoritm can not work well if the faces is not frontal face, I think.

  • Pic 4.
In [76]:
image_path = 'images/' + image_list[8]
predict_breed(image_path)
Dog and human were detected in this image
If this one were a dog, he / she would be a ... Golden_retriever!!
  • Pic 5.
In [77]:
image_path = 'images/' + image_list[-1]
predict_breed(image_path)
Human detector: True
If this one were a dog, he / she would be a ... Dogue_de_bordeaux!!
  • Pic 6.
In [78]:
image_path = 'images/' + image_list[9]
predict_breed(image_path)
Dog and human were detected in this image
If this one were a dog, he / she would be a ... Pembroke_welsh_corgi!!
  • Pic 7.
In [80]:
image_path = 'images/' + image_list[13]
predict_breed(image_path)
Dog and human were detected in this image
If this one were a dog, he / she would be a ... Welsh_springer_spaniel!!
  • Pic 8.
In [82]:
image_path = 'images/' + image_list[11]
predict_breed(image_path)
Dog and human were detected in this image
If this one were a dog, he / she would be a ... Pharaoh_hound!!
  • Pic 9.
In [83]:
image_path = 'images/dog_1.png'
predict_breed(image_path)
/opt/conda/lib/python3.6/site-packages/PIL/Image.py:931: UserWarning: Palette images with Transparency   expressed in bytes should be converted to RGBA images
  'to RGBA images')
Dog detector: True
This dog breed can be a Canaan_dog